To counter this and make XMPP work well on mobile and in browsers, we had to develop our own server and a bunch of new protocol extensions.
Real question: for email, I use K9, which supports IMAP Push, which, as far as I know, needs a live connection as well. Do you know if they are struggling as well, or is the xmpp push and imap push significantly different in some way?
Service -> Push service -> APNS / FCM -> Applications
With XMPP, servers act as 'service' from the scheme above, then relay a push initiation to 'Push service', maintained by a client app developers, Push service relays information to a respective service by Apple / Google, who deliver push to an Application, which connects to Service (or does not, pushes can contain all necessary information for displaying in notification, but that's a story for another time).
With email, I'm not really familiar with how IMAP Push works, but the scheme must be the same: an email server must relay info to a push service maintained by app developers, in your case, K9 guys.
From what I was able to gather after reading wikipedia for 5 min, an IMAP server upon receiving an email sends info to some 'mail user agent'. From wikipedia it's not really clear, but it appears that is is supposed to work on end devices. In this case, it won't really work as 'push'.
Google (or any other email provider, like Protonmail, who have their own app) get around this because they control both email server and push server, so it's trivial for them to initiate a push when an email is received.
A phone user expects that messages would be displayed to him in notifications when received. One does not expect to be cut off communications because he had pressed a 'home' button, but it's what happens, and a user does not have any control to prevent it from happening.
WhatsApp is a heavily modified xmpp, and does just that. That's what it works so well. Such approach is way easier in a federated environment.
Did you ever wonder why there are no (to date) XMPP clients on iOS that are not complete and utter shit?
As a person who has first-hand knowledge of the matter, managing a team member tasked with the development of an iOS XMPP app (which is supposed to be NOT complete and utter shit), I tell you that creation of such app is impossible without reinvention of half of XMPP stack (basically every part that touches client-server interaction). Thankfully, s2s works rather fine in XMPP.